Zero2: avoid graph breaks in torch.compile by using param_idx #6803
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
inside reduce_independent_p_g_buckets_and_remove_grads and in reduce_ipg_grads which are being executed during the BWD hook in zero2, the model param is being stored inside params_in_ipg_bucket. torch.compile has hard time tracing parameters. By using the param's static index inside the group the same logic can be maintain with less complexity.
tjruwase
reviewed
Dec 2, 2024
| @@ -1067,7 +1068,7 @@ def average_tensor(self, tensor): | |||
|
|
|||
| process_group = self.dp_process_group | |||
| # count = 0 | |||
| for i, param, param_id in self.params_in_ipg_bucket: | |||
| for i, param_idx_in_group, param_id in self.params_in_ipg_bucket: | |||
There was a problem hiding this comment.
Suggested change
| for i, param_idx_in_group, param_id in self.params_in_ipg_bucket: | |
| for i, param_idx_in_group, param_id in self.params_in_ipg_bucket: | |
| param = self.bit16_groups[group_idx][param_idx_in_group] |
param needs to be set as above for is_moe_param(param) in line 1075
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
inside reduce_independent_p_g_buckets_and_remove_grads and in reduce_ipg_grads which are being executed during the BWD hook in zero2, the model param is being stored inside params_in_ipg_bucket. torch.compile has hard time tracing parameters.
By using the param's static index inside the group the same logic can be maintain with less complexity.